Research On Optimization Of Gossip-based P2P Overlay Technology

P2P computing,as a new born technology in recent ten years,has attracted more and more attention.Its advantage is that making the best of users' resources for resource sharing,and then reducing the pressure of central servers.But as the network size increases,uncertainty of users increases and users can leave P2P overlay freely,which makes current large scale P2P overlay appears strong churn.How to ensure the scalability of P2P overlay but still maintaining the robustness,in another word,decreasing the negative effects of churn,is one of problems to be solved in the future P2P field.Gossip-based protocols mimic the feature of gossips spreading over people,and employ probability mechanism to disseminate information over networks,resulting in a good scalability and robustness.And combining gossip with P2P has been more and more popular during researchers.However,gossip-based P2P applications are still in junior period of development.There are three problems needs to be soloved when constructing large scale P2P applications:1)Random service peer selection leads to one-time wrong selection and unfair distribution of whole loads.2)Local peer adjusting policy leads to low abilities of superpeer filtering and low percentage controlling.3)Too many peers with low capability participate in mulicasting during large scale and dynamic environment leads to reliability and redundancy of multicast over P2P overlay.The dissertation's work is under support of National Scientific Important Project,"Research On Key Techonologies of Future Mobile Multimedia Video/Audio Codec"(No.2010ZX03004-003)and Scientific and Technological Project of Hubei Province,"FTTH Access Technology Stream"(No.2006AA101A03),does some research about service peer selection and unfair distribution of whole loads,low abilities of superpeer filtering and pencentage cotrolling,and reliability and redundancy during multicasting.Firstly the dissertation introduces background of P2P overlay and gossip-based P2P applications,then proposes a gossip-based P2P overlay model and gives related definitions.And directing at three problems mentioned above,the paper proposes load-balancing-oriented topology management protocol-LBTMP,adaptive hybrid P2P topology construction protocol-Firas,and highly reliable and lowly redundant multicast protocol over hybrid P2P overlay-DouLaCast.At last against the background of IPTV,the paper proposes a gossip-based design proposal for P2P-IPTV platform,and develops a prototype under proposal's guaidence.Finally the prototype has passed the acceptance check of Hubei High Technology Office.The main contributions of the dissertation are listed as follows:(1)Load-balancing based topology management protocol on P2P overlay LBTMP.Topology management,as a basic component in gossip-based large-scale P2P overlay,one of its functions is providing service peer selection for up-level P2P applications.Current topology management takes P2P overlay as a homogeneous overlay,using nodes' indegree as the metric for assessing the load-balancing of P2P overlay,and adopts random-based policy for service peer selection and topology information exchanging,without considering the differences among peers' ability,and variation of distribution of overlay loads,which leads to the probability of false drop ratio on service peers increasing more than 30%when loads increasing,and more unfair distribution of loads on P2P overlay,then exacerbating fairness of P2P overlay.Therefore,this paper does some research about loads model can describe by real time changing and metric of whole fairness of P2P overlay.We propose a model can describe real time variation of loads and improved P2P topology management protocol LBTMP.LBTMP can priorily select those who can afford more loads for service peers selection and topology information exchanging.What makes LBTMP can decrease the probability of false drop ratio and improve the distrition of loads on whole overlay.The simulations proves that no matter under static or dynamic scenarios,compared with random-based protocols,LBTMP can decrease FDR(False Drop Ratio)about 13%-18%,and decrease GLBR(Global Load-Balancing Ratio)about 14%-16%(smaller is better).(2)Distributed Global Adaptive Layered P2P Overlay Construction Protocol Firas.There is problem with lower capability of selecting superpeers and lower ratio controlling by abilities of superpeers when constructing a layered(Hybrid)P2P overlay.Current fiexed threshold partition method defines intervals in advance,and cannot adapt dynamic overlay environment.While current adaptive construction method only takes transition of peers' role through local neighbor list,whtat also makes low abilities of filtering superpeers and weak ratio controlling.These will result in low seach performance of hybrid P2P overlay.Therefor,this paper does some research about distributed peers grouping algorithm by peers'ability,and proposes adaptive hybrid P2P overlay construction protocol Firas according to the algorithm.Compared with traditional adaptive hybrid P2P overlay SG2,Firas carries out transition of peers' role not only by comparsion of local neighbor list,but mainly by bringing in cache of topology information and achieving Relative Postiion Value step by step in a global view,and finally improves accuracy of filtering superpeers and decreases error of ratio of superpeers controlling.It's useful for those hybrid P2P overlays who still don't achieve optimal status.However,Firas can control the transition of peers' role to avoid unnecessary transitions.The simulations valids that no matter in static or dynamic scenarios,compared with traditional SG2,Firas can increase the accuracy of superpeers filtering about 9%,and decrease error on ratio of superpeers controlling about 12%.(3)High Reliable and Low Redundancy Multicast Protocol DouLaCast on Hybrid P2P Overlay.Current gossip-based multicast protocol over P2P overlay takes probability mechanism to select forwarding peer for sending data,without considering the heterogeneity of P2P overlay,which will lead to massive lowly reliable and capable peers to participating in multicasting.It will increase the probability of accidence during probability-multicasting.Traditional multicast on hybrid overlay takes tree-based disseminating path or DHT structure.However,under large-scale dynamic environment router table in tree-based dissemination or DHT will be reconstructed frequently,producing disastrous influence to reliability and efficiency of multicast.Therefore,this paper combines hybrid overlay with multicast,proposing a hybrid mechanism multicast DouLaCast.It makes a mesh-based probability multicasting during superpeers on up-level,which can reduce lowly reliable and capable peers to participating in multicasting,and determinism multicasting to leafpeers on down-level,which can enhance reachability of data to lowly reliable peers.Compared with traditional flooding protocols,DouLaCast makes a tradeoff between reliability and communication redundancy.Moreover,DouLaCast brings in load-aware technique to adjust selection of forwarding peers,beneficial for load-balancing of P2P overlay.The simulations valids that under the same conditions in static scenarios,compared with the most reliable flooding protocols,DouLaCast has almost the same relibability but reduce more than 50%on traffic.And under the same conditions in dynamic scenarios,compared with HiGossip over hybrid overlay,DouLaCast improve 4%-9%on reliability.(4)Diverse applications-oriented P2P-IPTV platform model.Traditional IPTV system takes CDN as its dissemination model,what exists simplified dissemination model,weak scalability and low quality service.It cannot satisfy future high bandwidth and multi-service-oriented large-scale IPTV system.Therefore,this paper designs a gossip-based P2P-IPTV system model according to the idea of gossip-based large-scale P2P overlay.This model adopts a hybrid CDN-P2P dissemination method,carrying out CDN at the backbone of Internet,but constructing a hybrid P2P overlay for disseminating data and related data multicasting at the edge of Internet.Compared with traditional IPTV system,this platform model can quickly adapt the variation of environment,achieving robustness and scalability,resulting in strong support for large-scale applications and diverse service of IPTV.In conclusion,the work of this dissertation fully combines the advantages of gossip-based protocols and P2P overlay,exploring topology management,construction of topology and multicast in P2P overlay optimization field.It's of important theoriotical significance and application for large-scale dynamic P2P overlay applications.Finally we conclude the contributions of this paper and points out the future work.